Transcribe audio with the Grok Voice API

POST https://api.x.ai/v1/stt turns audio into text. Send multipart/form-data with either file or url (one is required). Put file last in the form. Get a key and credits at console.x.ai.

Unary call

curl -X POST https://api.x.ai/v1/stt \
  -H "Authorization: Bearer $XAI_API_KEY" \
  -F format=true \
  -F language=en \
  -F file=@audio.mp3

format=true needs language. That flag runs Inverse Text Normalization so spoken numbers come back in written form.

Limits and knobs

• Max file size: 500 MB.
• Containers auto-detected: wav, mp3, ogg, opus, flac, aac, mp4, m4a, mkv (MP3/AAC/FLAC inside mkv).
• Raw pcm / mulaw / alaw need audio_format plus sample_rate.
• keyterm biases proper nouns (repeatable, max 100 terms, 50 chars each).
• filler_words defaults to false (uh/um stripped).
• diarize=true adds a speaker label per word. multichannel=true returns channels.

Response shape

You get text, language (BCP-47), duration, and words as [{text, start, end}]. Optional speaker when diarize is on; optional channels when multichannel is on.

Streaming

Live audio uses wss://api.x.ai/v1/stt. Wait for transcript.created before sending binary frames, then finish with {"type":"audio.done"}. Never put the API key in the browser. Proxy the socket through your server.

Pitfalls

• 413 means the file is over 500 MB.
• 400 when format=true without language.
• 401 on a bad or missing key.
• Put file last. Fields after the file, or the file before other fields, may be ignored.
• Console API credits are a separate bill from SuperGrok's weekly pool on grok.com.